From e5c8d55530e4969716684465bfc90906fc572ea9 Mon Sep 17 00:00:00 2001
From: MichaelFisher1997 <contact@michaelfisher.tech>
Date: Sun, 3 May 2026 07:24:31 +0000
Subject: [PATCH] fix: retry observability oci chart fetches

---
 .gitea/workflows/deploy.yml                   | 23 ++++++++++++-------
 .../observability/ocirepository-loki.yaml     |  1 +
 .../observability/ocirepository-promtail.yaml |  1 +
 3 files changed, 17 insertions(+), 8 deletions(-)

diff --git a/.gitea/workflows/deploy.yml b/.gitea/workflows/deploy.yml
index 5f70379..8d03e12 100644
--- a/.gitea/workflows/deploy.yml
+++ b/.gitea/workflows/deploy.yml
@@ -967,17 +967,24 @@ jobs:
           wait_for_ocirepository_ready_or_cached() {
             local repository="$1"
             local timeout="$2"
+            local attempts="${3:-6}"
             local artifact_storage
+            local attempt
 
-            if kubectl -n flux-system wait --for=condition=Ready "ocirepository/${repository}" --timeout="${timeout}"; then
-              return 0
-            fi
+            for attempt in $(seq 1 "${attempts}"); do
+              reconcile_flux_resource "ocirepository/${repository}" 300
+              if kubectl -n flux-system wait --for=condition=Ready "ocirepository/${repository}" --timeout="${timeout}"; then
+                return 0
+              fi
 
-            artifact_storage="$(kubectl -n flux-system get "ocirepository/${repository}" -o jsonpath='{.status.conditions[?(@.type=="ArtifactInStorage")].status}' 2>/dev/null || true)"
-            if [ "${artifact_storage}" = "True" ]; then
-              echo "OCIRepository ${repository} is not currently Ready; continuing with cached artifact" >&2
-              return 0
-            fi
+              artifact_storage="$(kubectl -n flux-system get "ocirepository/${repository}" -o jsonpath='{.status.conditions[?(@.type=="ArtifactInStorage")].status}' 2>/dev/null || true)"
+              if [ "${artifact_storage}" = "True" ]; then
+                echo "OCIRepository ${repository} is not currently Ready; continuing with cached artifact" >&2
+                return 0
+              fi
+
+              echo "OCIRepository ${repository} did not become Ready after ${timeout}; forcing retry (${attempt}/${attempts})" >&2
+            done
 
             observability_diagnostics
             exit 1
diff --git a/infrastructure/addons/observability/ocirepository-loki.yaml b/infrastructure/addons/observability/ocirepository-loki.yaml
index 6a9bd98..96e7b2e 100644
--- a/infrastructure/addons/observability/ocirepository-loki.yaml
+++ b/infrastructure/addons/observability/ocirepository-loki.yaml
@@ -5,6 +5,7 @@ metadata:
   namespace: flux-system
 spec:
   interval: 10m
+  timeout: 5m
   url: oci://ghcr.io/grafana/helm-charts/loki
   ref:
     tag: 6.46.0
diff --git a/infrastructure/addons/observability/ocirepository-promtail.yaml b/infrastructure/addons/observability/ocirepository-promtail.yaml
index 064100c..1fb35f6 100644
--- a/infrastructure/addons/observability/ocirepository-promtail.yaml
+++ b/infrastructure/addons/observability/ocirepository-promtail.yaml
@@ -5,6 +5,7 @@ metadata:
   namespace: flux-system
 spec:
   interval: 10m
+  timeout: 5m
   url: oci://ghcr.io/grafana/helm-charts/promtail
   ref:
     tag: 6.16.6